On 2022/07/23 4:15, Fabio M. De Francesco wrote:
> I had misunderstood easily what you said because I had no context. I have
> not yet all the necessary context to prepare a "real" patch. As said, it
> was only a "proof of concept".
Although current problem is that iforce_usb_disconnect() being blocked on
dev->mutex which was held by the caller of iforce_close(), I worry that
a proper fix requires something like commit db264d4c66c0fe00 ("media:
imon: reorganize serialization") in order to defer
usb_free_urb(iforce_usb->irq);
usb_free_urb(iforce_usb->out);
kfree(iforce_usb);
part in iforce_usb_disconnect(), given that iforce_close() (which needs to
perform urb operation) might be called even after iforce_usb_disconnect()
completed.
Device file's close callback being not prepared for disconnect event
(that is, code assumes that close happens only before disconnect happens)
might be frequently observed problem in USB drivers...
